Building the Open Future of AI with Oleg Golev
Oleg Golev joins Titus to discuss the open-source AI movement, how open models change the security landscape, and why permissionless innovation matters for agent ecosystems.
They cover the tension between open weights and safety, composable agent architectures, and what builders should know about deploying open models in production. Oleg shares his perspective on why open-source AI is not just a philosophical preference but a practical necessity for building trustworthy agent systems.
Topics covered
- The case for open-weight models in production agent systems
- How open-source changes the security threat model
- Composable agent architectures and why modularity matters
- The role of identity and permissions in open agent ecosystems
- Practical advice for builders deploying open models today
- Where the open AI movement is headed in 2026 and beyond
Key takeaways
Open models need open guardrails. Oleg argues that the freedom to inspect and modify model weights is only valuable if you also have robust permission systems controlling what those models can do when deployed as agents.
Agent identity is the missing piece. The conversation keeps circling back to a fundamental question: how do you verify that an agent is who it claims to be, and how do you scope its permissions accordingly?